Using Machine Learning Techniques to Predict Viral Suppression Among People With HIV.

Area of Science:

• Health Informatics
• Machine Learning in Medicine
• Epidemiology

Background:

• Predicting viral suppression is crucial for managing HIV (Human Immunodeficiency Virus).
• Machine learning (ML) offers potential for improved predictive modeling in public health.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting viral suppression in people living with HIV (PWH) in South Carolina.
• To compare the performance of ML models against traditional statistical methods.

Main Methods:

• Utilized electronic health records from South Carolina (2005-2021) for adult PWH.
• Defined viral suppression as viral load <200 copies/mL.
• Employed Long Short-Term Memory Network and traditional models, analyzing data in 4-month windows with 1-, 3-, and 5-lagged time periods.

Main Results:

• The Long Short-Term Memory Network (LSTM) model demonstrated superior predictive performance (Lag 5 AUC = 0.881) compared to generalized linear mixed models.
• Key predictors included historical viral suppression, viral rebound, and viral blips.
• Incorporating county-level social vulnerability data did not enhance prediction accuracy.

Conclusions:

• Supervised machine learning algorithms, particularly LSTM, show promise for enhanced risk prediction of viral suppression in PWH.
• ML approaches may outperform traditional statistical methods in this clinical prediction task.